Merge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus
[linux-2.6] / fs / ocfs2 / stackglue.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * stackglue.c
5  *
6  * Code which implements an OCFS2 specific interface to underlying
7  * cluster stacks.
8  *
9  * Copyright (C) 2007, 2009 Oracle.  All rights reserved.
10  *
11  * This program is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU General Public
13  * License as published by the Free Software Foundation, version 2.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  */
20
21 #include <linux/list.h>
22 #include <linux/spinlock.h>
23 #include <linux/module.h>
24 #include <linux/slab.h>
25 #include <linux/kmod.h>
26 #include <linux/fs.h>
27 #include <linux/kobject.h>
28 #include <linux/sysfs.h>
29 #include <linux/sysctl.h>
30
31 #include "ocfs2_fs.h"
32
33 #include "stackglue.h"
34
35 #define OCFS2_STACK_PLUGIN_O2CB         "o2cb"
36 #define OCFS2_STACK_PLUGIN_USER         "user"
37 #define OCFS2_MAX_HB_CTL_PATH           256
38
39 static struct ocfs2_locking_protocol *lproto;
40 static DEFINE_SPINLOCK(ocfs2_stack_lock);
41 static LIST_HEAD(ocfs2_stack_list);
42 static char cluster_stack_name[OCFS2_STACK_LABEL_LEN + 1];
43 static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
44
45 /*
46  * The stack currently in use.  If not null, active_stack->sp_count > 0,
47  * the module is pinned, and the locking protocol cannot be changed.
48  */
49 static struct ocfs2_stack_plugin *active_stack;
50
51 static struct ocfs2_stack_plugin *ocfs2_stack_lookup(const char *name)
52 {
53         struct ocfs2_stack_plugin *p;
54
55         assert_spin_locked(&ocfs2_stack_lock);
56
57         list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
58                 if (!strcmp(p->sp_name, name))
59                         return p;
60         }
61
62         return NULL;
63 }
64
65 static int ocfs2_stack_driver_request(const char *stack_name,
66                                       const char *plugin_name)
67 {
68         int rc;
69         struct ocfs2_stack_plugin *p;
70
71         spin_lock(&ocfs2_stack_lock);
72
73         /*
74          * If the stack passed by the filesystem isn't the selected one,
75          * we can't continue.
76          */
77         if (strcmp(stack_name, cluster_stack_name)) {
78                 rc = -EBUSY;
79                 goto out;
80         }
81
82         if (active_stack) {
83                 /*
84                  * If the active stack isn't the one we want, it cannot
85                  * be selected right now.
86                  */
87                 if (!strcmp(active_stack->sp_name, plugin_name))
88                         rc = 0;
89                 else
90                         rc = -EBUSY;
91                 goto out;
92         }
93
94         p = ocfs2_stack_lookup(plugin_name);
95         if (!p || !try_module_get(p->sp_owner)) {
96                 rc = -ENOENT;
97                 goto out;
98         }
99
100         active_stack = p;
101         rc = 0;
102
103 out:
104         /* If we found it, pin it */
105         if (!rc)
106                 active_stack->sp_count++;
107
108         spin_unlock(&ocfs2_stack_lock);
109         return rc;
110 }
111
112 /*
113  * This function looks up the appropriate stack and makes it active.  If
114  * there is no stack, it tries to load it.  It will fail if the stack still
115  * cannot be found.  It will also fail if a different stack is in use.
116  */
117 static int ocfs2_stack_driver_get(const char *stack_name)
118 {
119         int rc;
120         char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
121
122         /*
123          * Classic stack does not pass in a stack name.  This is
124          * compatible with older tools as well.
125          */
126         if (!stack_name || !*stack_name)
127                 stack_name = OCFS2_STACK_PLUGIN_O2CB;
128
129         if (strlen(stack_name) != OCFS2_STACK_LABEL_LEN) {
130                 printk(KERN_ERR
131                        "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
132                        stack_name);
133                 return -EINVAL;
134         }
135
136         /* Anything that isn't the classic stack is a user stack */
137         if (strcmp(stack_name, OCFS2_STACK_PLUGIN_O2CB))
138                 plugin_name = OCFS2_STACK_PLUGIN_USER;
139
140         rc = ocfs2_stack_driver_request(stack_name, plugin_name);
141         if (rc == -ENOENT) {
142                 request_module("ocfs2_stack_%s", plugin_name);
143                 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
144         }
145
146         if (rc == -ENOENT) {
147                 printk(KERN_ERR
148                        "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
149                        plugin_name);
150         } else if (rc == -EBUSY) {
151                 printk(KERN_ERR
152                        "ocfs2: A different cluster stack is in use\n");
153         }
154
155         return rc;
156 }
157
158 static void ocfs2_stack_driver_put(void)
159 {
160         spin_lock(&ocfs2_stack_lock);
161         BUG_ON(active_stack == NULL);
162         BUG_ON(active_stack->sp_count == 0);
163
164         active_stack->sp_count--;
165         if (!active_stack->sp_count) {
166                 module_put(active_stack->sp_owner);
167                 active_stack = NULL;
168         }
169         spin_unlock(&ocfs2_stack_lock);
170 }
171
172 int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
173 {
174         int rc;
175
176         spin_lock(&ocfs2_stack_lock);
177         if (!ocfs2_stack_lookup(plugin->sp_name)) {
178                 plugin->sp_count = 0;
179                 plugin->sp_proto = lproto;
180                 list_add(&plugin->sp_list, &ocfs2_stack_list);
181                 printk(KERN_INFO "ocfs2: Registered cluster interface %s\n",
182                        plugin->sp_name);
183                 rc = 0;
184         } else {
185                 printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
186                        plugin->sp_name);
187                 rc = -EEXIST;
188         }
189         spin_unlock(&ocfs2_stack_lock);
190
191         return rc;
192 }
193 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
194
195 void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
196 {
197         struct ocfs2_stack_plugin *p;
198
199         spin_lock(&ocfs2_stack_lock);
200         p = ocfs2_stack_lookup(plugin->sp_name);
201         if (p) {
202                 BUG_ON(p != plugin);
203                 BUG_ON(plugin == active_stack);
204                 BUG_ON(plugin->sp_count != 0);
205                 list_del_init(&plugin->sp_list);
206                 printk(KERN_INFO "ocfs2: Unregistered cluster interface %s\n",
207                        plugin->sp_name);
208         } else {
209                 printk(KERN_ERR "Stack \"%s\" is not registered\n",
210                        plugin->sp_name);
211         }
212         spin_unlock(&ocfs2_stack_lock);
213 }
214 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
215
216 void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto)
217 {
218         struct ocfs2_stack_plugin *p;
219
220         BUG_ON(proto == NULL);
221
222         spin_lock(&ocfs2_stack_lock);
223         BUG_ON(active_stack != NULL);
224
225         lproto = proto;
226         list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
227                 p->sp_proto = lproto;
228         }
229
230         spin_unlock(&ocfs2_stack_lock);
231 }
232 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_locking_protocol);
233
234
235 /*
236  * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take
237  * "struct ocfs2_lock_res *astarg" instead of "void *astarg" because the
238  * underlying stack plugins need to pilfer the lksb off of the lock_res.
239  * If some other structure needs to be passed as an astarg, the plugins
240  * will need to be given a different avenue to the lksb.
241  */
242 int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
243                    int mode,
244                    union ocfs2_dlm_lksb *lksb,
245                    u32 flags,
246                    void *name,
247                    unsigned int namelen,
248                    struct ocfs2_lock_res *astarg)
249 {
250         BUG_ON(lproto == NULL);
251
252         return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
253                                               name, namelen, astarg);
254 }
255 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
256
257 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
258                      union ocfs2_dlm_lksb *lksb,
259                      u32 flags,
260                      struct ocfs2_lock_res *astarg)
261 {
262         BUG_ON(lproto == NULL);
263
264         return active_stack->sp_ops->dlm_unlock(conn, lksb, flags, astarg);
265 }
266 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
267
268 int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
269 {
270         return active_stack->sp_ops->lock_status(lksb);
271 }
272 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
273
274 int ocfs2_dlm_lvb_valid(union ocfs2_dlm_lksb *lksb)
275 {
276         return active_stack->sp_ops->lvb_valid(lksb);
277 }
278 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
279
280 void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb)
281 {
282         return active_stack->sp_ops->lock_lvb(lksb);
283 }
284 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
285
286 void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
287 {
288         active_stack->sp_ops->dump_lksb(lksb);
289 }
290 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
291
292 int ocfs2_stack_supports_plocks(void)
293 {
294         return active_stack && active_stack->sp_ops->plock;
295 }
296 EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
297
298 /*
299  * ocfs2_plock() can only be safely called if
300  * ocfs2_stack_supports_plocks() returned true
301  */
302 int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
303                 struct file *file, int cmd, struct file_lock *fl)
304 {
305         WARN_ON_ONCE(active_stack->sp_ops->plock == NULL);
306         if (active_stack->sp_ops->plock)
307                 return active_stack->sp_ops->plock(conn, ino, file, cmd, fl);
308         return -EOPNOTSUPP;
309 }
310 EXPORT_SYMBOL_GPL(ocfs2_plock);
311
312 int ocfs2_cluster_connect(const char *stack_name,
313                           const char *group,
314                           int grouplen,
315                           void (*recovery_handler)(int node_num,
316                                                    void *recovery_data),
317                           void *recovery_data,
318                           struct ocfs2_cluster_connection **conn)
319 {
320         int rc = 0;
321         struct ocfs2_cluster_connection *new_conn;
322
323         BUG_ON(group == NULL);
324         BUG_ON(conn == NULL);
325         BUG_ON(recovery_handler == NULL);
326
327         if (grouplen > GROUP_NAME_MAX) {
328                 rc = -EINVAL;
329                 goto out;
330         }
331
332         new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
333                            GFP_KERNEL);
334         if (!new_conn) {
335                 rc = -ENOMEM;
336                 goto out;
337         }
338
339         memcpy(new_conn->cc_name, group, grouplen);
340         new_conn->cc_namelen = grouplen;
341         new_conn->cc_recovery_handler = recovery_handler;
342         new_conn->cc_recovery_data = recovery_data;
343
344         /* Start the new connection at our maximum compatibility level */
345         new_conn->cc_version = lproto->lp_max_version;
346
347         /* This will pin the stack driver if successful */
348         rc = ocfs2_stack_driver_get(stack_name);
349         if (rc)
350                 goto out_free;
351
352         rc = active_stack->sp_ops->connect(new_conn);
353         if (rc) {
354                 ocfs2_stack_driver_put();
355                 goto out_free;
356         }
357
358         *conn = new_conn;
359
360 out_free:
361         if (rc)
362                 kfree(new_conn);
363
364 out:
365         return rc;
366 }
367 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
368
369 /* If hangup_pending is 0, the stack driver will be dropped */
370 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
371                              int hangup_pending)
372 {
373         int ret;
374
375         BUG_ON(conn == NULL);
376
377         ret = active_stack->sp_ops->disconnect(conn);
378
379         /* XXX Should we free it anyway? */
380         if (!ret) {
381                 kfree(conn);
382                 if (!hangup_pending)
383                         ocfs2_stack_driver_put();
384         }
385
386         return ret;
387 }
388 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
389
390 /*
391  * Leave the group for this filesystem.  This is executed by a userspace
392  * program (stored in ocfs2_hb_ctl_path).
393  */
394 static void ocfs2_leave_group(const char *group)
395 {
396         int ret;
397         char *argv[5], *envp[3];
398
399         argv[0] = ocfs2_hb_ctl_path;
400         argv[1] = "-K";
401         argv[2] = "-u";
402         argv[3] = (char *)group;
403         argv[4] = NULL;
404
405         /* minimal command environment taken from cpu_run_sbin_hotplug */
406         envp[0] = "HOME=/";
407         envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
408         envp[2] = NULL;
409
410         ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
411         if (ret < 0) {
412                 printk(KERN_ERR
413                        "ocfs2: Error %d running user helper "
414                        "\"%s %s %s %s\"\n",
415                        ret, argv[0], argv[1], argv[2], argv[3]);
416         }
417 }
418
419 /*
420  * Hangup is a required post-umount.  ocfs2-tools software expects the
421  * filesystem to call "ocfs2_hb_ctl" during unmount.  This happens
422  * regardless of whether the DLM got started, so we can't do it
423  * in ocfs2_cluster_disconnect().  The ocfs2_leave_group() function does
424  * the actual work.
425  */
426 void ocfs2_cluster_hangup(const char *group, int grouplen)
427 {
428         BUG_ON(group == NULL);
429         BUG_ON(group[grouplen] != '\0');
430
431         ocfs2_leave_group(group);
432
433         /* cluster_disconnect() was called with hangup_pending==1 */
434         ocfs2_stack_driver_put();
435 }
436 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
437
438 int ocfs2_cluster_this_node(unsigned int *node)
439 {
440         return active_stack->sp_ops->this_node(node);
441 }
442 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
443
444
445 /*
446  * Sysfs bits
447  */
448
449 static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
450                                                struct kobj_attribute *attr,
451                                                char *buf)
452 {
453         ssize_t ret = 0;
454
455         spin_lock(&ocfs2_stack_lock);
456         if (lproto)
457                 ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
458                                lproto->lp_max_version.pv_major,
459                                lproto->lp_max_version.pv_minor);
460         spin_unlock(&ocfs2_stack_lock);
461
462         return ret;
463 }
464
465 static struct kobj_attribute ocfs2_attr_max_locking_protocol =
466         __ATTR(max_locking_protocol, S_IFREG | S_IRUGO,
467                ocfs2_max_locking_protocol_show, NULL);
468
469 static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
470                                                  struct kobj_attribute *attr,
471                                                  char *buf)
472 {
473         ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
474         struct ocfs2_stack_plugin *p;
475
476         spin_lock(&ocfs2_stack_lock);
477         list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
478                 ret = snprintf(buf, remain, "%s\n",
479                                p->sp_name);
480                 if (ret < 0) {
481                         total = ret;
482                         break;
483                 }
484                 if (ret == remain) {
485                         /* snprintf() didn't fit */
486                         total = -E2BIG;
487                         break;
488                 }
489                 total += ret;
490                 remain -= ret;
491         }
492         spin_unlock(&ocfs2_stack_lock);
493
494         return total;
495 }
496
497 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
498         __ATTR(loaded_cluster_plugins, S_IFREG | S_IRUGO,
499                ocfs2_loaded_cluster_plugins_show, NULL);
500
501 static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
502                                                 struct kobj_attribute *attr,
503                                                 char *buf)
504 {
505         ssize_t ret = 0;
506
507         spin_lock(&ocfs2_stack_lock);
508         if (active_stack) {
509                 ret = snprintf(buf, PAGE_SIZE, "%s\n",
510                                active_stack->sp_name);
511                 if (ret == PAGE_SIZE)
512                         ret = -E2BIG;
513         }
514         spin_unlock(&ocfs2_stack_lock);
515
516         return ret;
517 }
518
519 static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
520         __ATTR(active_cluster_plugin, S_IFREG | S_IRUGO,
521                ocfs2_active_cluster_plugin_show, NULL);
522
523 static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
524                                         struct kobj_attribute *attr,
525                                         char *buf)
526 {
527         ssize_t ret;
528         spin_lock(&ocfs2_stack_lock);
529         ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
530         spin_unlock(&ocfs2_stack_lock);
531
532         return ret;
533 }
534
535 static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
536                                          struct kobj_attribute *attr,
537                                          const char *buf, size_t count)
538 {
539         size_t len = count;
540         ssize_t ret;
541
542         if (len == 0)
543                 return len;
544
545         if (buf[len - 1] == '\n')
546                 len--;
547
548         if ((len != OCFS2_STACK_LABEL_LEN) ||
549             (strnlen(buf, len) != len))
550                 return -EINVAL;
551
552         spin_lock(&ocfs2_stack_lock);
553         if (active_stack) {
554                 if (!strncmp(buf, cluster_stack_name, len))
555                         ret = count;
556                 else
557                         ret = -EBUSY;
558         } else {
559                 memcpy(cluster_stack_name, buf, len);
560                 ret = count;
561         }
562         spin_unlock(&ocfs2_stack_lock);
563
564         return ret;
565 }
566
567
568 static struct kobj_attribute ocfs2_attr_cluster_stack =
569         __ATTR(cluster_stack, S_IFREG | S_IRUGO | S_IWUSR,
570                ocfs2_cluster_stack_show,
571                ocfs2_cluster_stack_store);
572
573 static struct attribute *ocfs2_attrs[] = {
574         &ocfs2_attr_max_locking_protocol.attr,
575         &ocfs2_attr_loaded_cluster_plugins.attr,
576         &ocfs2_attr_active_cluster_plugin.attr,
577         &ocfs2_attr_cluster_stack.attr,
578         NULL,
579 };
580
581 static struct attribute_group ocfs2_attr_group = {
582         .attrs = ocfs2_attrs,
583 };
584
585 static struct kset *ocfs2_kset;
586
587 static void ocfs2_sysfs_exit(void)
588 {
589         kset_unregister(ocfs2_kset);
590 }
591
592 static int ocfs2_sysfs_init(void)
593 {
594         int ret;
595
596         ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
597         if (!ocfs2_kset)
598                 return -ENOMEM;
599
600         ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
601         if (ret)
602                 goto error;
603
604         return 0;
605
606 error:
607         kset_unregister(ocfs2_kset);
608         return ret;
609 }
610
611 /*
612  * Sysctl bits
613  *
614  * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path.  The 'nm' doesn't
615  * make as much sense in a multiple cluster stack world, but it's safer
616  * and easier to preserve the name.
617  */
618
619 #define FS_OCFS2_NM             1
620
621 static ctl_table ocfs2_nm_table[] = {
622         {
623                 .ctl_name       = 1,
624                 .procname       = "hb_ctl_path",
625                 .data           = ocfs2_hb_ctl_path,
626                 .maxlen         = OCFS2_MAX_HB_CTL_PATH,
627                 .mode           = 0644,
628                 .proc_handler   = &proc_dostring,
629                 .strategy       = &sysctl_string,
630         },
631         { .ctl_name = 0 }
632 };
633
634 static ctl_table ocfs2_mod_table[] = {
635         {
636                 .ctl_name       = FS_OCFS2_NM,
637                 .procname       = "nm",
638                 .data           = NULL,
639                 .maxlen         = 0,
640                 .mode           = 0555,
641                 .child          = ocfs2_nm_table
642         },
643         { .ctl_name = 0}
644 };
645
646 static ctl_table ocfs2_kern_table[] = {
647         {
648                 .ctl_name       = FS_OCFS2,
649                 .procname       = "ocfs2",
650                 .data           = NULL,
651                 .maxlen         = 0,
652                 .mode           = 0555,
653                 .child          = ocfs2_mod_table
654         },
655         { .ctl_name = 0}
656 };
657
658 static ctl_table ocfs2_root_table[] = {
659         {
660                 .ctl_name       = CTL_FS,
661                 .procname       = "fs",
662                 .data           = NULL,
663                 .maxlen         = 0,
664                 .mode           = 0555,
665                 .child          = ocfs2_kern_table
666         },
667         { .ctl_name = 0 }
668 };
669
670 static struct ctl_table_header *ocfs2_table_header = NULL;
671
672
673 /*
674  * Initialization
675  */
676
677 static int __init ocfs2_stack_glue_init(void)
678 {
679         strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
680
681         ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
682         if (!ocfs2_table_header) {
683                 printk(KERN_ERR
684                        "ocfs2 stack glue: unable to register sysctl\n");
685                 return -ENOMEM; /* or something. */
686         }
687
688         return ocfs2_sysfs_init();
689 }
690
691 static void __exit ocfs2_stack_glue_exit(void)
692 {
693         lproto = NULL;
694         ocfs2_sysfs_exit();
695         if (ocfs2_table_header)
696                 unregister_sysctl_table(ocfs2_table_header);
697 }
698
699 MODULE_AUTHOR("Oracle");
700 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
701 MODULE_LICENSE("GPL");
702 module_init(ocfs2_stack_glue_init);
703 module_exit(ocfs2_stack_glue_exit);